Skip to content

ToolsEnabled Sim 0.1.7 (beta)

Pre-release
Pre-release

Choose a tag to compare

@ToolsEnabled-Inc ToolsEnabled-Inc released this 13 Oct 12:36

ToolsEnabled Sim 0.1.7 (beta) explains how to run several robots on one OS account with Claude Code, gives every MCP tool a top-level title, and makes a missing browser fail with the same clear message in every render test, while preserving the pinned physics engines and recorded trajectory bytes. MIT licensed.

What changed

  • Several robots on one OS account. Claude Code stores the plugin's connection_file once per user, in settings.json of its configuration directory, not per project: installing for a second robot from another project rebinds the first, and pluginConfigs in a project's settings files is ignored. Give each robot agent its own CLAUDE_CONFIG_DIR and use it for every claude command of that agent (or pass a --settings file on every launch). README and docs/CLAUDE-INSTALL.md show the commands.
  • Expected install message. "1 userConfig option not yet set" refers to the optional run_root; the install succeeded. Add --config run_root=/absolute/private-sim-run to set it as well.
  • Top-level tool titles. Every MCP tool now has a top-level title (MCP 2025-06-18) equal to its annotation title. Hints are unchanged: sim.observe, sim.status and sim.assignments are read-only; the nine tools that change the shared world are marked destructive; no tool reaches beyond your computer.
  • One browser check. All render tests use the same offline Chromium check as capture, so a missing browser always fails with "Offline Chromium missing. Set SIM_CHROMIUM_EXECUTABLE…" instead of a raw launcher error.
  • Install details. For a Git-backed install, Claude Code runs npm ci --ignore-scripts in the installed plugin directory itself; run it by hand only if node_modules is missing. claude plugin details shows Hooks (0) because it counts only event hooks in hooks/hooks.json; Sim loads its mod through that file's modules list.
  • Physics unchanged. Dependency versions and vendored bytes are unchanged: Rapier 0.17.3, MuJoCo 3.14.0, Three.js 0.180.0 and playwright-core 1.62.0.

Get started

  1. Download toolsenabled-sim-0.1.7.zip and toolsenabled-sim-0.1.7.zip.sha256, then run sha256sum -c toolsenabled-sim-0.1.7.zip.sha256.
  2. With Node.js 22 or later, run unzip toolsenabled-sim-0.1.7.zip, cd sim, node tools/demo.mjs --output ./runs/demo, then node tools/replay.mjs ./runs/demo. For MuJoCo: node tools/demo.mjs --output ./runs/demo-mujoco --engine mujoco, then node tools/replay.mjs ./runs/demo-mujoco.
  3. To connect your own agents, see docs/LIVE.md: node tools/live.mjs --root /absolute/run --work-record /absolute/work-record.json (add --engine mujoco for MuJoCo). See docs/CLAUDE-INSTALL.md for the marketplace commands, one CLAUDE_CONFIG_DIR per robot agent, and per-agent private connection options.

Tested on this exact release (source commit e382434101397d906b5f1f1350262c94f510f94a, tag v0.1.7; ZIP SHA-256 30aba253a98a7e00af7f98b81e23293877d23fe0e11024400ee422c9e1ce3150; Linux, Node.js 22.19.0, Claude Code 2.1.296)

  • 116 of 116 Node tests pass headless on source, including actual Chromium rendering and a test that lists every tool through the real server and pins its top-level title, annotation title and hints. All 14 packaging/scanner tests and the DeepSeek configuration test pass on source. The 44 plugin hook tests and strict Claude validation pass on source and on the extracted ZIP; strict validation also passes on plugin.json and the marketplace catalog.
  • The ZIP rebuilds byte for byte twice from this clean commit: 5,439,107 bytes, 141 files under one sim/ wrapper.
  • From the extracted ZIP, the scripted demo delivers three crates with no collisions and replays byte-identically over 600 ticks on both engines. Trajectory SHA-256: Rapier 453aa41446485c2865437652a7c4a6256fb395b8ccf52dc7ae6ec35f4c8e53a5; MuJoCo 278254654043d0aa09299940eddb56eef083a0377fd0597868ba2cb10b7ea9d6, the same as 0.1.6.
  • A local Claude Code marketplace add and install from this commit lists toolsenabled-sim 0.1.7. With one CLAUDE_CONFIG_DIR per robot, two configurations bound to the amber and teal connection files each showed the server Connected in claude mcp list and started it with its own robot's file. Real model-driven sessions, native Desktop installation and Codex's Plugins Directory installation remain separate integration qualifications.

Known limits

  • Linux was tested. Native Windows is unsupported because Sim relies on POSIX file ownership checks. Native Claude Desktop installation and production bundle signing need separate qualification; no MCPB is attached.
  • A Sim action starts only from your own prompt, at the terminal or through Remote Control. A headless claude -p run, a scheduled or relayed message, another session or another plugin cannot start one.
  • A Git-backed install takes the repository's default branch at its current commit. Use the verified extracted ZIP for this exact release and offline dependencies. Catalog registration alone does not establish pane or MCP readiness.
  • The pinned Rapier compatibility initializer prints an upstream notice on stderr once per process. Recorded physics bytes and MCP stdout are unaffected.
  • Seven dependency files exceed 256 KiB; THIRD-PARTY.md lists the shipped files and licenses. No browser binary ships. Identical rendered frames require the same Chromium build and renderer.
  • The handoff scanner cannot inspect keys hidden in raw deflate data without a recognizable header.
  • One-tick live runs at the default recording cap have a practical ceiling of about 30,000 ticks; larger scenes or heavier traffic can stop sooner. The planner does not predict other robots' future paths; the coordinator schedules crossing traffic.
  • Driver labels are supplied by the operator. Processes running as the same OS user can read private files and act as that user. See SECURITY.md. Security policy: https://toolsenabled.ai/security/; support: support@toolsenabled.ai.

ToolsEnabled is not affiliated with or endorsed by Anthropic, OpenAI, DeepSeek, Cursor, Google DeepMind or any other AI provider.